Chapter 2 Rhythmic Patterns in Language It

Chapter 2 Rhythmic Patterns in Language

21 Generalities

Speech-like music that other culturally codified production of sound-is a rhythmically organized activity Just as the rhythmic schemes of the different musical traditions are a codification presumshyably reflecting both universal properties of rhythmic organization and the conventionalized choices about particulars made by the culture (from the range of choices that are available within the confines of a (universally defined) rhythmic organization) so too we will argue do the diverse rhythmic patterns of prominence found in language reflect both universal and particular aspects of rhythmic organization The rhythmic patterning of every language is codified and the linguistic description of speech rhythm therefore is a description of that code A successful theory of rhythmic structure in language will involve sorting out what is language-particular and what is universal in that code

Rhythm as we said in chapter 1 is founded upon a recurrence of pulses some of which emerge as accented with respect to each other forming patterns on different levels Various aspects of rhythmic orshyganization inhere in the metrical grid representation A grid position is a pulse a point in (abstract) time The notion of pattern requires a differentiation among the pulses and this is provided by ascribing to them a place on different levels Sequences of rhythmically equivalent pulses are represented as the sequence ofgrid positions on the different metrical levels To every level then there is a patterning Thus claiming that speech is rhythmically organized- that in the linguistic representation of an utterance syllables are aligned with a metrical grid-amounts to claiming that (a) there are discernable pulses in speech (b) some of these syllables emerge as accented with respect to

Generalities 37

each other forming recurrent patterns and (c) there may be a hierarshychy of accented syllables corresponding to patterns on the different metrical levels It may be argued and has been that all these stateshyments hold true for speech

What does not inhere in the grid in our conception of it is the preshycise nature of the patterns themselves An accented (strong) position could in principle be followed or preceded by any number of nonshyaccented (weak) positions This we believe is entirely appropriate the grid itself simply represents patterns as the organization of pulses of abstract time into levels The patterns are characterized by principles or rules that govern the particulars of the metrical grid organization

There is arguably a universal rhythmic ideal one that favors a strict alternation of strong and weak beats Following Sweet 1875-76 we will call this as in chapter 1 the Principle of Rhythmic Alternation (PRA) That the rhythmic organization oflanguage or music aspires to such an ideal state is indicated by a number of general tendencies that are attested in rhythmic patterning (i) binary patterns (successions of ws or sw) at various levels are by far the most prevalent in either domain of activity (ii) ternary patterns (wws or sww) are not (usually) basic in the rhythmic patterning at any level but rather exist alongside binary patterns and constitute a special departure from them and quaternary patterns simply do not exist inasmuch as they may be inshyterpreted as two binary patterns This is Sweets point one also implicit in Liberman 1975 In his work on rhythm in French Dell (to appear) also argues for a universal ideal rhythmic organization following Ie principe d ewythmie according to which an ideal metrical grid contains no adjacent strong positions and is maximally alternating

Our position is not that the PRA plays a direct role in the linguistic description of a languages patterns of prominence but rather that the rules of a grammar that define the possible metrical grid alignments of the sentences of that language conspire in approximating that ideal on every level of organization At the end of this chapter we will outline our theory of metrical grid alignment (or theory of the notion possible rhythmic score for language) and will show how we think the rhythmic ideal is embodied in the rules for constructing a metrical grid on the basis of a text In the immediately following section we will examine the evidence for viewing speech as a rhythmically organized activity one that both respects an abstract rhythmic structure having the character of a metrical grid alignment and shows patterns reflec ting the operation of the PRA

38Rhythmic Patterns in Language

22 The Rhythmic Nature of Speech

Scholars working on the phonetics ofEnglish have long recognized that rhythmic organization at what we will call the lower levels plays an important role in the description of speech 1 A major insight of these investigations is that the quantities (durations) of the syllables of a senshytence are determined by the sentences rhythmic properties According

to D Jones 1964sect886 for example Vowel length depends to a considerable extent [in English on the rhythm of the sentence There is a strong tendency in connected speech to make stressed syllables follow each other as nearly as possible at equal distances For D Jones 1964sect888-890 the musical notation in (2la) and (21b) represents the fact that in (21a) ei ai are something like twice as long in the second sequence as they are in the first and that in (21b) the i in scene is considerably longer than the i in scenery

(21) a n

eitin oi~ FJ

naintin twenti vs )

eit ) )

nain ten

eighteen nineteen twenty eight nine ten

b )i ) m )rnr ()~ sin w~z bjUltafi vs 6a sinari waz bjutafl The scene was beautiful The scenery was beautiful

Pike 194534 makes a not dissimilar observation about American Enshyglish He points out that the space of time elapsing between the stressed syllables man and here is roughly the same in The man is here and The managers here with the result that in the latter the syllables are crushed together (pronounced very rapidly) in order to be fitted

in Building on the insights of Jones and others Abercrombie 1964 laid

the foundation for further work oil the topic by positing a unit of analyshysis the foot (not to be confused with the foot of metrical phonology) in terms of which he argued the rhythmic properties of syllables may be

explained English utterances may be considered as being divided by the isoshychronous beat of the stress pulse into feet of (approximately) even

The Rhythmic Nature of Speech 39

length Each foot starts with a stress and contains everything that folshylows that stress up to but not including the next stress This is the house that lack built has therefore four feet and they can be most conveniently represented by the use of vertical lines

This is thelhouse thatlJacklbuilt The quantity of any syllable is a proportion of the total length of the foot within which the syllable occurs and it is relative to the quantity of any other syllable in the foot (1964217)

The claim being made then is that English utterances can be charshyacterized in rhythmic terms as consisting of a sequence of isochronous pulses This claim about isochrony in English speech has engendered considerable debate (see Lehiste 1980 for a review) 2 We feel that much of this debate and the apparent lack of resolution on the question stems from a failure to understand rhythm and the isochrony of its basic pulses in sufficiently abstract terms With Liberman Lehiste and others we claim a psychological reality for rhythmic organization in both the production and the perception of speech but acknowledge that the ideal isochronicity this organization presupposes may not alshyways reveal itself in easily measurable terms in the acoustic signal

We may liken the metrical grid alignment of a sentence- the linguisshytic representation of the sentences rhythmic structure-to a musical score It must be recognized that a musical score or a metrical grid alignment though grounded in its very conception by the rhythmic cashypacities of the human organism (the rules for defining possible scores or metrical grid alignments embodying as they do universals of rhythmic organization) is but an abstract scheme The same score may be intershypreted in many different ways by the same or different performers though of course there are limits within which variation in interpretashytion is confined if the score is to remain recognizable that is if the score is to be considered to have been realized in the performance A distinction must therefore be made between the score and its interpreshytation This is the distinction between langue and parole or between competence and performance (using the latter term now in its technical sense) For linguistic patterns of prominence this is the distinction between the metrical grid alignment of the sentence and its phonetic implementation In measuring for isochrony one is measuring only the performance of the score and not the abstract patterning that makes it up We give full credence therefore to the impression of isochronyshyto the impression of rhythm-as revealing something about how the mind grasps the organization of speech in time 3 It is what the mind

Rhythmic Patterns in Language 40

grasps that is of interest to us here for that is presumably what is emshybodied in the metrical grid alignment of a sentence Indeed isochrony should in no way be considered as the sine qua non of a truly rhythmic system The notion of pattern as consisting of a regular recurrence of motifs defined in strong-weak terms may be just as important as isoshychrony in establishing the rhythmic character of speech

The thrust of the observations about isochrony in English is that not all syllables are rhythmically equivalent to each other only some are Moreover only some are in the relation of (ideal) isochrony with reshyspect to each other This means that all syllables do not qualify as pulses in Cooper and Meyers sense Rather it is only the stressed Abercrombian foot-initial syllables that align with the pulses or beats of the rhythmic organization of English speech The others have in fact a somewhat variable realization in time depending in part on how many of them there are between the basic pulses What does this mean for the alignment of these syllables with the metrical grid We could contemplate giving Pikes second example the grid alignment below where the sequence of xs represents the lowest level of the metrical grid (and the higher levels of beats are not represented)

(22) x x

The managers here

Here only the syllables that coincide with a pulse an x are specified for an alignment The others are in limbo so to speak Alternatively we could postulate a yet lower level on the metrical grid one with which all syllables would be aligned but not one where the points are taken to mark out ideally isochronous pulses Given this approach the pulses or beats would be represented only on the second level and above The managers here would have the following grid alignment (still incomshyplete at the higher levels)

(23) x x

x x x x x The managers here

As will become clear it is extremely useful to assume that all syllables of the utterance are integrated into the rhythmic organization of the utterance in the sense of having a specified alignment with the metrical grid In particular this assumption makes possible a straightforward


treatment of patterns of word stress a description of rather fine details of timing within feet and a representation of subtle aspects of syntactic timing We will therefore adopt the latter sort of metrical grid alignment for language Within the metrical grids of speech rhythm we will thus distinguish a first metrical level (the lowest) a second metrical level and any number of levels (in principle infinitely many) above that The positions on the second metrical level and above will be referred to as beats The positions on the second metrical level will be referred to more specifically as basic beats Finally the positions on the first level will be referred to as demibeats this term being chosen to reflect the only quasi-pulse status of the positions on the lowest level

It may be argued that the difference between languages that Pike 1945 has called stress-timed like English and those that are syllableshytimed can and indeed must be represented at the basic beat level of rhythmic organization A syllable-timed language is one in which it is said that there is a tendency towards isochrony of all syllables a relashytive constancy in the duration of syllables in the utterance and a (relashytive) lack of vowel reduction (Pike 194535-37 Catford 197785-88 Abercrombie 196796-98) French Italian and Spanish are often cited as examples of syllable-timed languages 4 For example in the Italian if popolo the syllables are pronounced as a staccato progression of evenly spaced beats Viewed in metrical terms a syllable-timed lanshyguage is one in which each syllable is aligned with a (basic) beat in the metrical grid Given the assumption that beats appear only at the second metrical level then the minimal well-formed syllable-to-grid alignment for if popolo will be as follows

(24)

x x xx x x x x il popolo

Evidence for this particular conception of the grid alignment for syllable-timed languages is presented in Selkirk (in preparation)

So far then we have demonstrated some reason for positing two levels of rhythmic organization in speech the level of basic beats and a lower level of demibeats It is around these levels that the major part of the discussion concerning isochrony in speech has revolved s It is also at these levels that the most copious evidence concerning rhythmic patterning has accumulated as for example in recent work on word stress where it has been shown that in stress-timed languages the nurnshy


ber of weak stressless syllables intervening between stressed sylshylables is usually one or two when there is a pattern at all (see Halle and Vergnaud 1979 Safir 1979 Hayes 1980) In other words in the patterns of stress that are attested in natural language it appears that at most two weak demibeats intervene between strong ones This sort of syllashyble count gives obvious support to our formulation of the Principle of Rhythmic Alternation and more generally to the idea that speech is a rhythmically organized activity

In studies subsequent to Abercrombie 1964 such as Catford 1966 and Halliday 1967a the rhythmic foot has been seen as the central and in fact unique unit in the analysis of sentential rhythm6 Translated into the terms in which we have been discussing the problem of rhythm this amounts to saying that only two metrical levels are involved in speech rhythm syllables are organized into basic beats and thats it If this were indeed a correct assessment of the facts the motivation for inshyvoking the complexly hierarchized representation of rhythm that is embodied in the metrical grid would be weak However as we will demonstrate speech does show a greater hierarchy of rhythmic arshyrangements Indeed the termfoot as defined by Abercrombie is releshyvant only to these lower levels of rhythmic organization where the difference between stress-timing and syllable-timing is represented

What of the degrees of rhythmic prominence (often referred to in the tradition of Trager and Smith 1951 as degrees of stress)-that is the distinctions between strong and weak beats on various levels It seems that these distinctions have been recognized but that in the notational representation of rhythm they have been obscured Catford for examshyple allows the sentence John bought two books last week to have any of the following arrangements into feet

(25)

John I bought I two I books I last I week John I bought two I books last I week

I John bought I two books I last week John bought two I books last week etc

But the notion foot employed here cannot be the same one defined by Abercrombie for each of the monosyllabic words in these examples is stressed and so constitutes a foot on its own The vertical marks must in fact be taken as indicating strong-weak relations on a level above that of the foot The point is made clear when multisyllabic


words consisting of a stressed syllable followed by unstressed syllables are substituted for the monosyllables in (25) for example Mary purshychased twenty pamphlets yesterday morning The intuition is that the same placements of vertical lines are appropriate that is that the same rhythmic groupings are possible for the sentence Yet within the spans of the utterance flanked by the lines still further rhythmic distinctions are made as shown below where the italics indicate local rhythmic prominences (beats)

(26)

I Mary Ipurchased I twenty I pamphlets I yesterday I morning Mary Ipurchased twenty Ipamphlets yesterday I morning

I Mary purchased I twenty pamphlets I yesterday morning Mary purchased twenty Ipamphlets yesterday morning etc

The intuition then is that rhythmic groupings are made at more than one level7 This along with other evidence (to be reviewed directly) that there are degrees (or levels) of rhythmic prominence in speech shows that a hierarchical representation of speech rhythm such as the metrical grid is necessary

The necessity for distinguishing a minimum of two metrical levels above the basic beat level in the metrical grid with which a sentence is aligned is quite common Consider the English sentence Abernathy gesshytieulated Here some syllables are aligned with beats (they are marked with accents) and some are not and among the beat-aligned syllables some are aligned with a strong beat (they bear acute accents) Thus in this example a strong-weak alternation of beats is clearly perceived Any representation of speech rhythm requires some means of denoting this strong-weak contrast The alignment of the words of the sentence with the metrical grid would minimally involve three metrical levels (27) x x x x x x x x x x x xxxx Abernathy gesticulated

The syllables aligned with strong beats here (coinciding with points on the third metrical level) are often referred to as syllables bearing word stress or more specifically main word stress And indeed in the words of many languages whether they are stress-timed like English or sylshy


lable-timed like Italian there will be a locus of rhythmic prominence a beat reliably stronger than the others The existence of (main) word stress then indicates rhythmic organization above the basic beat level

The strong beat of main word stress does not mark the highest level of rhythmic organization in an English sentence or in the sentences of many other languages In the normal English sentence there exists a strong-weak distinction between beats at more than one level a beat that is strong on one level may coincide with a beat on a higher level that is either weak or strong This is the case in the sentence Abernathy gesticulated where the syllable -ti- is the most prominent of all either in the neutral pronunciation of the sentence with a pitch accent only on the verb or in the nonneutral pronunciation in which both words bear pitch accents The syllable -ti- may be said to bear main sentence or phrase stress The full alignment of the sentence with the grid would be as follows

(28)

x x x x x x x x x x x x xxxx Abernathy gesticulated

Thus the existence of just one degree of sentence or phrase stress alongside main word stress shows that at least two levels of rhythmic organization must be distinguished above the level of basic beats Again it is quite common among languages for the main-stressed syllashybles of the words making up a sentence to differ in their degrees of rhythmic prominence Partly the existence and location of phrasal rhythmic prominence are to be attributed to the operation of rules of grammar that are sensitive to syntactic structure such as the Nuclear Stress Rule of English and partly the appearance of rhythmic promishynence on the phrase is to be attributed simply to the demands of rhythmic organization per se and in particular to the PRA The differshyentcontributions to phrasal prominence will be discussed in section 23 and in chapters 3 and 4

One potentially very telling sort of nonintuitional evidence for deshygrees of rhythmic prominence in speech is hinted at by Pierrehumbert 1980 She shows that the phonetic values for the tones (ie pitch acshycents) composing intonational contours in English are in part a function of the relative prominence of the syllables with which they are asso-


ciated where prominence for Pierrehumbert is largely (though not entirely) a matter of stress ie rhythmic prominence8 Pierrehumbert offers a convincing case for representing intonational contours as conshysisting (largely) of a sequence of atomic pitch accents She argues that these pitch accents which associate with the main stresses of words are to be characterized either as one of two single level tones high (H) and low (L) or as binary combinations of these (see section 53) Moreover she shows that in the same intonational contour given a pair of pitch accents consisting say of one high tone each the high tone that is associated with the rhythmically most prominent syllable will consistently have the higher frequency (when the effects of declishynation are factored out of course)9 This relation holds not only beshytween a nuclear pitch accent and a prenuclear pitch accent but also between two prenuclear pitch accents with different prominence valshyues as shown in figure 21 In this example In November the regions weather Was unusually dry each of the principal words bears a high tone pitch accent The sentence has a declarative contour It corresponds to two intonational phrases demarcated here with the symbol (For the particulars of such contours see section 53)

The English Nuclear Stress Rule determines that weather will be more prominent than regions that dry will be more prominent than unusually and that the most prominent element of the last phrase (ie dry) will be more prominent than any other element of the sentence The alignment of the sentence with the metrical grid would thus be minimally as shown in figure 2210 A comparison of the two figures shows them to be consistent with Pierrehumberts suggested generalshyization that Fo height is a reflection of relative rhythmic prominence We will pursue this question of the relation between intonation and rhythm in chapters 4 and 5 For now we seek only to point out that there exists evidence beyond intuitions about timing that points to the sort of rhythmic organization being claimed to exist

Liberman 1975 and Liberman and Prince 1977 have offered another important sort of evidence from English for degrees of rhythmic promishynence and specifically for representing these prominence relations in terms of levels on the metrical grid They argue that to properly charshyacterize the conditions under which the pervasive phenomenon of stress shift (often referred to as the Rhythm Rule) takes place crushycially requires making a distinction in levels of rhythmic organization Consider for example the location of stress prominences in the normal pronunciation of the phrases below 11


M~v CL L

I IN NOVEMBER THE REGIONS WEATHER WAS UNUSUALLY DRY

I I I I I HL L H HI H HLL

Figure 21

x x X

X X X X X XX X X X X X

X X X X X XX X X X X XX X X

In November the regions weather was unusually dry

Figure 22

(29) Dundee marmalade (the) thirteenth of May Westminster Abbey (the) unknown soldier (a) good-looking lffeguard anaphoric reference thirteen men achromatic lens

(The acute accent indicates primary word stress and the grave indishycates secondary word stress in the terms used here strong vs weak beats) Of significance is the fact that main stress in the first word is not located where it would be if that word were pronounced in isolation or in other phrasal contexts Dundee thlrteen imaphOric more goodshylooking than you etc The main stress has been shifted backward in the examples above Liberman and Princes approach to this matter is to say roughly that a surface representation like achromatic Lens is derived by a rule of stress shift from an underlying representation in which the first word has its normal lexical word stress pattern eg achromatic Lens Liberman and Prince argue that the stress shift rule applies only when two stresses of the same level are adjacent in the relevant sense and hence constitute a stress clash The stress clash is eliminated by shifting the first of the clashing stresses backward

Clearly strict adjacency of stressed syllables is not required as a preshycondition for stress shift for the rule applies in the phrases with Westminster good-looking and anaphoric as it does in (the) thirteenth


(of May) where a stressless syllable intervenes between the two apparshyently clashing stressed syllables Thus as Liberman and Prince point out the offending adjacency must be represented as occurring at a higher level it cannot be explained purely in terms of the linear arshyrangement of syllables This notion of higher level can of course be characterized in terms of the metrical grid or in other not strictly linear representations of stress 12

Assume that before stress shift the metrical grid alignment of achromatic lens is as shown in (2 lOa) and that stress shift converts it into (21Ob)

(210)

a X b x X ~ ~

X X x

x x

X X xx x x x xx x achromatic lens achromatic lens

(The alignment at the first three levels in (2 lOa) is determined by the principles of word stress in English and the fourth level beat on lens corresponds to the effects of the Nuclear Stress Rule) It is easy to explain in terms of the grid why the concatenation of rhythmic units in (2 lOa) is ill formed and therefore subject to modification by stress shift In general a grid configuration of the follOWing sort will be judged deshyviant regardless of what metrical levels are involved (211)

x x

x x

This is because the grid (ie the rhythmic structure)fails to alternate in such a configuration in the sense that two strong beats are not sepashyrated by a weak beat This offending configuration appears in the repshyresentation (21Oa) ofachromatic lens where it is circled Thus though the stressed syllables in (2 lOa) are not themselves adjacent the strong beats with which they are aligned are in the metrical grid and this is what constitutes the stress clash Thus the clash can be defined here only in terms of the second metrical level and above Supporting this characterization of the stress clash is the fact that stress shift does not take place as readily or at all When the main stress is some distance away and thus not involved in a clash We will see examples of this in


section 43 where the Rhythm Rule of English is discussed in some detail

As Liberman and Prince point out of course stress shift is not the only means available for avoiding a stress clash in the grid Alternashytively the final syllable of the first word may be lengthened or a pause may be placed between the words It may be argued that the posshysibilities of lengthening and pausing are attributable to the presence of additional silent beats whose appearance is governed by syntactic factors (see section 43 and chapter 6) and which themselves prevent stress clash from arising

What is important to the more general point at hand is the fact that characterizing the ill-formed grid configuration referred to as a clash requires reference to two metrical levels In the particular cases under discussion it means that reference is made to both the basic beat level and one level above and this in tum means that stress clash and its consequent stress shift provide evidence for the existence of at least three levels of rhythmic organization in English words

In discussing rhythmic organization in speech and in particular the metrical grid we have suggested that language exhibits a quite general Principle ofRhythmic Alternation (PRA) which determines the patterns of alternation appearing in the grid A provisional formulation of the PRA is as follows

(212) Between two successive strong beats on a metrical level n there must intervene at least one (and at most two) weak beat(s) of the metrical level n

Note now that in (211) the two points on the lower metrical level are both strong beats in that they both coincide with beats of the next metrical level up and that because the two strong beats are adjacent this configuration does not conform to the PRA It seems reasonable then to view the PRA as in some way responsible for ruling the stress clash in (2 lOa) ill formed and this is in fact the view we will take (see section 23) Note that although the PRA itself does not refer to more than one metrical level (the strong and weak beats to which it refers are all of one level) the very notions of strong beat and weak beat that it invokes require reference to two metrical levels Thus the charshyacterization of a stress clash as a configuration ruled ill formed by the PRA does involve an appeal to more than one metrical level


The characterization of stress clash given by Liberman 1975 and Liberman and Prince 1977312-313 does not invoke a general principle of alternation Instead the offending ill-formedness is defined directly as that grid configuration in which two positions on level m are not separated by a position on the next lower level m -1 Either approach makes the case for higher levels of rhythmic organization and the two are equivalent for the cases examined so far As we will show howshyever there is quite independent motivation for the PRA-it apparently operates in numerous other circumstances where the notion of stress clash is simply irrelevant The value of characterizing the ill-formedshyness of a stress clash in the way advocated here then is that this analshyysis invokes a principle of some generality in the grammar not one relevant only to the problem at hand

Another sort of arhythmicity may arise in the derivation of the phoshynological representation of a word or sentence one that like the stress clash is often enough done away with in the surface phonological repshyresentation of a sentence This other sort of grid configuration is an overlong sequence of weak beats that is not punctuated by any strongas depicted in (213)

(213) 000

xxx

(The symbol 0 indicates the absence of any beat at the point that 0

occupies in the grid) Here too we find an absence of rhythmic altershynation We might call this a rhythmic lapse to distinguish it from the clash It will be desirable to (re)formulate the PRA so that it can rule out both types of nonalternating configuration

Considerable evidence shows that the rhythmic organization of speech abhors a lapse as much as it does a stress clash We can see this in the rhythmic organization of phrases For example for a normal neutral pronunciation of sentence (214) where there is a pitch acshycent only on the final word which also bears main phrase stress (see sections 422 and 55) there must be a rhythmic prominence before the final main stress

(214)

(I know quite well that) its organized on the model of a gallon of worms


The metrical grid alignment (215) is avoided in favor of some alternatshying pattern be it that of (216) or (217)13 (To simplify matters the lowest metrical level noted here corresponds to primary word stress

(Le the third metrical level))

(215) x x x x x

its organized on the model of a gallon of worms

(216) x xx

x xx x

(217) x

x x x x x x (215) is the minimal grid alignment consistent with the Nuclear Stress Rule (NSR) and the assignment of a pitch accent to the final word (see chapter 4) But consistency with the NSR and pitch accent assignment is not enough to make the metrical grid alignment well formed if the grid does not otherwise exhibit the appropriate alternation in rhythmic organization Something like the PRA would seem to be at work

Another set of facts indicating that the rhythmic organization of speech avoids lapses involves secondary word stress In both stressshytimed and syllable-timed languages the rules of the grammar will define the alignment of syllables with basic beats and often enough will also pick out which of these beats is the most prominent (main-stressed) in the word For example the rules for basic beat alignment and main word stress in syllable-timed Italian will give the grid alignment (218)

for the penultimately stressed word generativa

(218) x

x x x xx x x x xx

generativa But this alignment which involves a lapse preceding the main stress is not actually attested What is found instead is a secondary stress preshyceding the main stress at a distance of either one or two syllables


(Malagoli 1946 Nespor and Vogel 1979 Chierchia 1982b Vogel and Scalise 1982) as illustrated in (219) and (220)

(219) (220) x x

x x x x x x x x x x x xx x x x xx x x x xxx

generativa generativa

We will also be proposing (section 23 and chapter 3) that the very existence of alternation at the lowest (basic beat and demibeat) levels is to be ascribed to the PRA in some way A sequence of syllables each of which is aligned only with a (weak) demibeat at the first metrical level is but one very long rhythmic lapse and is obviously avoided

In general the tendency to avoid lapses at all levels of rhythmic organization is overwhelming We will reformulate the PRA to reflect this As provisionally stated in (212) the PRA rules out any instance of clash as well as those instances of lapse where strong beats (or demishybeats) flank a sequence of more than two weaks But there are other instances of rhythmic lapse where the overlong sequence of weaks is not flanked on both sides by strongs eg (215) and (218) If the failshyure of these configurations to appear in surface phonetic representation is to be attributed to the PRA then it must be reformulated so as not to require the presence of the flanking strongs Suppose the PRA specified instead that any weak beat or demibeat may be preceded by at most one other weak This says in effect that there may be at most two weaks in sequence It has the effect of ruling out overlong weak seshyquences between strong beats or demibeats and such sequences not flanked on both sides as well Let us call this the anti-lapse provision of the PRA

We might now consider reformulating the anti-clash provision of the PRA which currently states that at least one weak beat or demibeat must intervene between two strongs In fact one alternative to this appeal to the flanking strongs appears to make certain correct predicshytions about the facts that the earlier formulation leaves unexplained This alternative consists in specifying simply that at least one weak beat (or demibeat) must follow a strong14 Like the first this formulashytion ensures that there will be at least one weak between two strongs unlike the first however it ensures in addition that a weak beat (or demibeat) will always follow a final strong beat (The notion final will be defined with respect to some domain -word phrase etC-and


thus is not limited to sentence-final position) For the time being then let us entertain the formulation in (221) as an alternative to (212)

(221) The Principle ofRhythmic Alternation

a Every strong position on a metrical level n should be followed by at least one weak position on that level

b Any weak position on a metrical level n may be preceded by at most one weak position on that level

The consequences of formulating the principle in this way will be exshyamined more fully in the following chapters

To sum up then there appears to be considerable evidence for the rhythmic organization of speech and for representing that rhythmic organization as an alignment of the syllables of a sentence with posishytions in a metrical grid Like a musical score the alignment of a senshytence with a grid represents the (ideal) isochrony of the pulses of speech the relative durations of individual syllables and their degrees of relative prominence Moreover the grid permits an understanding of the alternations in the rhythmic realizations of words in the sentence This is an extremely important point The metrical grid allows the genshyeralization to be expressed that the same rhythmic ideal the Principle of Rhythmic Alternation governs (in a way yet to be defined) the patterns to be found at allevels The PRA formulated with respect to levels of the metrical grid expresses the propensity to alternation at the lower levels in the same terms as the propensity to avoid clash or to introshyduce alternation at the higher levels Thus viewing stress patterns in terms of the metrical grid rationalizes the properties of stress systems even further than the metrical tree theory of stress patterns for it ratioshynalizes the pattern question itself (see sections 122 and 42) It is for this reason that we view the metrical grid as crucial to a theory ofstress patterns and thereby more than worthy of supplanting the now otiose trees in this function

23 Building the Grid

231 The Framework We are interested here in developing a theory of patterns of rhythmic prominence in language a theory of the notion possible metrical grid alignment We believe that that theory must incorporate universals of

Building the Grid 53

rhythmic organization and must delineate the range of choices that are available to grammars of individual languages for the codification of particular rhythmic patterns What we are aiming for is a core theory of rhythmic patterns (see Chomsky 1981 and references cited therein) Given such a core theory the grammar of an individual language will specify not language-particular rules but which among the (universally defined) rules made available by the theory are actually at play in a particular language Following Chomsky we will call this a languageshyparticular specification of the parameters delineated by the theory

In recent years considerable research has been done in metrical phoshynology with a view toward developing a core theory of word stress patterns based on a metrical (prosodic) tree representation of promishynence relations This effort has been advanced most notably by the work of Halle and Vergnaud 1979 and developed more recently in Hayes 1980 As will become clear many of the fundamental insights into the parameters of stress theory gained in these works and others are readily and perspicuously characterized in a metrical grid frameshywork With Prince 1981 1983 we argue that a theory of stress patterns is better expressed when the metrical grid is assumed to be the basic representation of stress Many of the basic lines of thinking pursued here find their inspiration in Princes work though the particular arshyticulations of the ideas are different in certain respects having been developed independently

As stated earlier we propose that the relation between the text (a syntactic representation) and its metrical grid alignment is to be exshypressed as a set of rules that construct a metrical grid alignment for the text according to the principle of the cycle On this theory these rules progressively build up the metrical grid alignment of a sentence (a rhythmic score) from the lower levels to the higher on successively larger cyclic domains An alternative theory is entirely conceivable according to which the text is in one way or another matched up with a full-fledged grid somehow independently defined a grid that may undergo modification once aligned with the text This is the approach implicit in Liberman 1975 Liberman and Prince 1977 and Dell (to apshypear) There the text which includes the abstract stress pattern of the sentence is matched up with a preexistent grid without appeal to the cycle Let us consider this alternative theory stripped of its asshysumption that the text includes metrical tree patterns of stress

Two issues must be distinguished here whether the text-grid relation is established by construction or matching and whether the relashy


tion is established cyclically or not We submit that the principle of the cycle is essential to the proper characterization of this relation and we will argue for this position in several ways in this chapter and in chapshyters 3 and 7 The fact that the text-grid relation must be cyclically defined removes any possibility of entertaining the matching theory we believe Fragments of a metrical grid corresponding to the lowest cyshyclic domains could perhaps be considered to be preexistent and matched with the text But there could no longer be a preexistent metshyrical grid of the entire sentence in any interesting sense given that the grid matched on lower cycles would be modified cyclically And to say that the grid is only matched at the lowest levels removes any force from this proposal making it very much like the construction theory It will become evident that a grid construction model permits the various sorts of generalizations about patterns of rhythmic prominence that a theory must capture to emerge quite perspicuously at all levels of the grid and it is for this reason that we adopt such a model here

The theory of metrical grid construction has two major components The first is a set of text-fo-grid alignment (TGA) rules Through these rules the particular properties of the text may impose requirements on the rhythmic realization of the sentence For example it may be stated in the grammar of a language as a TGA rule that syllables having such and such a degree of sonority must receive rhythmic prominence while syllables of lower sonority do not Or it may be stated that syllables aligned with beats located at the beginnings or ends of words or phrases must receive more rhythmic prominence than others in the overall rhythmic structure of the word or phrase Thus a number of important properties of particular stress systems follow not so much from the properties of the grid itself as from the principles in the grammar of individual languages that govern just how the syllables of their uttershyances (texts) are aligned with the grid Quite certainly one of the major descriptive tasks for a metrical grid theory of stress is to characterize the TGA rules available to language As we will show it seems that these rules fall into just four classes (i) the (universal) rule of Demishybeat Alignment (DBA) which aligns each syllable with a single grid position on the lowest metrical level (ii) the basic beat rules which align syllables with beats on the second metrical level by virtue of (a) their composition (ie the composition of their rime) andor (b) their position with respect to a particular syntactic domain (iii) the domainshyend prominence rules (the End Rules of Prince 1983) which ensure the greatest prominence of syllables aligned with beats at the beginning or


end of some specified syntactic domain by promoting them to beat hood on metrical level three or higher (eg the English Main (Word) Stress Rule Compound Rule and Nuclear Stress Rule) and (iv) the Pitch Accent Prominence Rule (PAR) which ensures that syllables with which the pitch accents forming an intonational contour are aligned are rhythmically prominent with respect to non-pitch-accented syJlables These rule types will be examined in later chapters

The Principle of Rhythmic Alternation defines an ideal rhythmic orshyganization by requiring that a strong beat be followed by a weak and that a weak be preceded by at most one other weak it defines an ideal metrical grid Yet the rules of language for aligning syllables with a grid-the TGA rules-pay no heed to this ideal From the point of view of the PRA these rules are capable of producing chaos or more precisely undesirable lapses and clashes Nonetheless the PRA is a sort of Platonic ideal to which the rhythmic structure grounded in sylshylables tones and syntactic structure aspires It is the rules of grid euphony (GE) that aid actual sentences in attaining this ideal and they thus form the second major component of the core theory of promi nence patterns We will give evidence for three types of grid euphony rules (i) Beat Addition (ii) Beat Movement and (iii) Beat Deletion Rules of these types attempt to set things right they build rhythmic order or some semblance of it out of the chaos that the TGA rules are capable of producing We gather them together here 15 16

(222)

Beat Addition

x a x x x x (left-dominant addition)

x b x x x x (right-dominant addition)

(223)

Beat Movement

x x x x x x

a x x x x x x (left movement) x x x x x x

b x x x x x x (right movement)


(224) Beat Deletion

x x x x x c x x x

a x x = x x xx=xx x x x x x d x x x

b x x = x x xx=xx

The interesting property of these rules is that they are defined solely in terms of the grid make no reference to the properties of the text and apply in principle at any metrical level They are purely rhythmic in character In our view they are the collective essence of the PRA One important set of parameters to be fixed in the grammar of an indishyvidual language will be the indication of whether or not one or more of these rule types applies on a given metrical level and a specification of which particular rule laquoa) (b) etc) the language chooses In other words just how the PRA is realized in the rhythmic structure of a lanshyguage and to what extent it is is a matter for language-particular deshyscription The universal of rhythmic organization in language then is that by means ofgrid euphony rules languages aspire in one way or another to meet the PRA in their metrical grid alignments

A fundamental question now arises How do the GE rules interact with TGA rules like the Nuclear Stress Rule the domain-end promishynence rules of word stress and so on More specifically under what circumstances may the GE rules obliterate the effects of the TGA rules The answer makes crucial use of the notion of the cycle We will show that within a cyclic domain the GE rules may not undo the prominence relations that are required on that domain by TGA rules and that there is reason to posit the following general condition on grid construction

(225) Textual Prominence Preservation Condition (TPPC)

A text-to-grid alignment rule applying on a syntactic domain d is necessarily satisfied on that domain

This principle has far-reaching implications for a theory of metrical grid alignments For example it will ensure that on a given cyclic domain Beat Addition may not change the location of a TGA-specified greatest prominence on that domain so that it will not undo the effects of the Main Stress Rule in words or the Nuclear Stress Rule on phrases on


that domain but only complement them This is the correct result as we will show

232 The Interaction of the Two Grid Construction Components

2321 The First Metrical Level It seems reasonable to entertain the hypothesis that universally phonological representations conform to this principle every syllable of an utterance is aligned with at least one demibeat in the metrical grid This principle simply says that every syllable participates in the rhythmic organization of the utterance At this point in our investigation it would be counterproductive to assume otherwise The principle as stated is probably not restrictive enough however since it would allow a single syllable to be aligned with sevshyeral beats at once But such a situation does not seem to arise except in the rather limited circumstances involving syntactic timing 17 For exshyample at the limits of words and phrases a final syllable that is aligned underlyingly with a single beat or demibeat may in addition be aligned in a derived representation with one or more silent grid posishytions whose presence in the grid is attributable to the syntactically governed silent demibeat addition that gives rise to syntactic timing (see chapter 6) We propose then as a condition on the initial text-toshygrid alignment (the first step in the construction of the metrical grid preceding silent demibeat addition) that a syllable align with just one demibeat Under the grid construction approach this is stated as the following (universal) TGA rule

(226) Demibeat Alignment

Align just one demibeat with every syllable

This rule would derive the alignment of (227b) from the syllable seshyquence (227a) on the lowest cyclic domain

(227) a cr cr cr cr cr cr b x x x x x x

I I I I cr cr cr crcr cr

This formulation of Demibeat Alignment coupled with the restrictive assumption that the other TGA rules do not augment the grid horizonshy


tally by adding grid positions on the first metrical level will ensure that except in circumstances created by the silent demibeat addition of syntactic timing there will be one demibeat for every syllable and one syllable for every demibeat We will call this additional assumption about TGA rules the Minimal Demibeat Condition It may tum out to be too restrictive an assumption in particular as far as the basic beat rules are concerned18 but we will take it as a part of our working hyshypothesis for the moment

2322 The Second Metrical Level If the grammar of a language conshytained no rules other than Demibeat Alignment for governing the alignment of the syllables of the text with the grid but if it did allow Beat Addition to apply on the second metrical level and above then that language would be a stress-timed language of a rather special sort The alignment of syllables with basic beats and beats on higher levels would be determined neither by the nature of the syllables themselves nor by their position in the word or the phrase but simply by the rhythmic tendencies expressed in Beat Addition The patterns of beats might vary freely in their location within words In fact we know of no language like this The grammars of languages with which we are acshyquainted often do include additional rules that in conjunction with Demibeat Alignment and GE rules determine in more specific fashion how syllables are to be aligned with basic beats of the second metrical level These are the basic beat rules

A syllable-timed language has only one basic beat rule which specifies that every syllable will have a basic beat (see Selkirk (in prepashyrationraquo) We could formulate it provisionally as align every syllable with a beat But in stress-timed languages at least those of the more commonplace sort whether or not a syllable is aligned with a basic beat seems to be able to depend on its composition its position in the doshymain both of these factors or neither (as in the hypothetical case cited above) We will suppose that universal grammar makes available the following rule schemata and that languages may choose their basic beat rules from these

(228) Basic Beat Rules

a Align a syllable of compositional type x with a beat

b Align a syllable in position y with a beat


A syllable-timed language falls under the universal rule schema (22Sa) in such a language every syllable is of type x A theory of stressshytimed languages will involve a theory of (i) what distinctions in comshypositional type may be appealed to in basic beat rules and (ii) what sorts of specifications of position within the cyclic domain may be made We will not attempt to elaborate such a theory here We repeat however the widely made observation that the possibilities of stressshying a syllable seem to involve only the properties of the rime CODshy

stituent of the syllable not those of the onset (Pike and Pike 1947 Kurylowicz 1948) The position has been taken that the relevant propshyerties of the rime are to be expressed in geometrical terms-in other words that the branching of a syllable into one or more constituents is what is crucial in determining its place within stress patterns 19 Howshyever there is good reason to believe that pure geometry is not the imshyportant factor It is known that the syllable type distinctions that playa role in attested stress systems may have to do with the quantity or weight of a syllable (Le whether or not it has a long vowel or whether or not it has a long vowel or is closed by a consonant) or the quality of the vowel contained in the syllable or the syllables tonal specifications 2o Little insight is to be gained by seeing the last distincshytion as geometric Prince 1981 1983 hypothesizes that they are all ultimately to be explained as distinctions in the sonority of the syllashyble21 As for position the cases with which we are familiar involve domain-initial position domain-final position or the position before the main stress (strongest beat) of a word 22 If these cases are representashytive then the class ofpositions is quite narrowly circumscribed Again though it is not our purpose at present to pursue the theory of word stress from the point of view of the syllable types or positions involved in stress rules for these issues are not crucially related to the conshycerns of this study

Our interest is in understanding the patterns that result from the presence in the grammar of basic beat rules invoking these type and position distinctions Note that these language-particular basic beat rules do not mention an alternating pattern of any sort Our hypothesis is that the attested alternations in weak and strong demibeats are in a sense to be made precise largely the contribution of GE rules and in particular the rule(s) of Beat Addition It is in defense of this hypotheshysis that we hope to establish the well-foundedness of the metrical grid theory of stress


Given a grid construction approach the basic beat rules are instrucshytions for adding beats to the grid on the second metrical level A rule of type (b) in which y stands for at the beginning of the word would read

(229) Align a (basic) beat with the first syllable on the domain Word

It would derive the alignment in (230) from that in (227b) supposing that sequence to be a Word

(230) x x x x x x x

(J (J (J (J (J (J

In this way Demibeat Alignment and the language-particular basic beat rules jointly construct a partial grid A partial grid so produced is not necessarily a well-formed rhythmic structure however (230) for exshyample contains a rhythmic lapse and is therefore abhorrent to the PRA Within the grid construction framework we are proposing that partial grid is further built up by the pattern-engendering GE rules The main agent of this build-up we suggest is a rule that adds beats-the rule of Beat Addition

For most non-syllable-timed languages Beat Addition is obligatory at the basic beat level (see section 325) which is to say that most such languages exhibit some sort of regular alternating pattern at that level There may possibly be some languages that have no Beat Addition on the second metrical level 23 but more typically it seems languages will require that Beat Addition apply on the basic beat level Specifically they will require that either (222a) or (222b) apply Adapting a term borrowed from Hayes 1980 to our own uses we will call Beat Addition (222a) left-dominant and Beat Addition (222b) right-dominant This then is one parameter the grammar of a language may set There is yet another to be set for Beat Addition at the basic beat level It is well known that patterns of stressed and stressless syllables may be estabshylished in directional fashion from right to left (R-L) or from left to right (L-R) across some particular domain In terms of the present analysis this means that Beat Addition on the second metrical level may be directional in nature the two values of that parameter being R-L and L-R Notice now that the obligatoriness of Beat Addition


and its directional character ensure that spans of the domain not disshyturbed by the presence of a basic beat established by TGA rules will exhibit a strictly alternating binary pattern This is precisely what is found in stress-timed languages and therefore confirms our general apshyproach to the matter

Suppose that a language has no basic beat rule(s) but does have obligatory left-dominant Beat Addition applying from right to left Such a specification of parameters would produce the patterns in (231)

(231) odd even

x x x x

x x x

x x x x x x x x x

These are the patterns for odd and even syllables attested in Warao (Osborn 1916 cited in Hayes 1980) Other parameter settings in other languages will give a basic beat on the final syllable combined with alternation from right to left and so on

Consider in this light the hypothetical partial grid (230) In a lanshyguage requiring no Beat Addition it would stay as it is But four other completions of the second level are possible when Beat Addition comes into play which arise as a result of specifying the parameters for domishynance and directionality The language Maranungku (Tryon 1970 cited by Hayes 1980) is one in which the domain-initial basic beat rule (228b) is at play In this language the initial syllable is stressed as is every other odd syllable including the final one in odd-syllabled words Given the present framework this indicates that Maranungku has a right-dominant rule of Beat Addition that proceeds left to right The output of this Beat Addition is shown in (232)

(232) odd even

x x x x x x x x x x x xxxxxx

Let us consider the example of a language with a basic beat rule involving the compositional syllable type namely Cairene Classical Arabic Cairene Classical Arabic is the pronunciation of classical Arabic used by speakers of the Cairene dialect as reported by Mitchell 1960 McCarthy 1979ab offers an extremely plausible account of the stress pattern of this dialect drawing on Mitchells description Two


aspects of the stress pattern are of interest here First heavy syllables (CVC or CVV) must be feet on their own to use McCarthys terms Second a sequence of nonheavy syllables enters into a (preferentially) binary pattern of beats going from left to right and starting from the beginning of the word and from any heavy syllable The first stressed syllable in a sequence of light syllables following a heavy syllable is located immediately after the heavy

(233)

CV CV CV CVV CV CV CV CVC

McCarthyS analysis of Cairene Classical patterns at the foot level translates into the following analysis of basic beats in the metrical grid framework

(234) a Basic Beat Rule

Every heavy syllable is aligned with a basic beat

b Beat Addition

Left-dominant Left to right

Thus the beat patterns of the words kaataba inkasara and adshywiyatuhu will be derived as follows

(235) kaataba inkasara adwiyatuhu

DBA t t t x x x x xxx x x x x x

kaataba inkasara adwiyatuhu BBR t t t

x x x x x x x xxx x x x x x

kaataba inkasara adwiyatuhu

BA t t t x x x x x x x x x x x xxx x x x x x


The last syllable in inkasara is not promoted to basic beat status by the left-dominant Beat Addition simply because its structural description is not met 24


With these brief examples we do not purport to have shown the superiority of a metrical grid theory of patterns of stressed syllables such as the one we have proposed It is beyond the scope of this book to attempt this We refer the reader first to Prince 1983 who has made a persuasive case for a metrical grid theory of these patterns covering the range of phenomena that has been treated in earlier metrical tree analyses and second to chapter 3 of this volume which will demonmiddot strate the appeal of a metrical grid approach to English word stress In sketching the preceding analyses we have simply attempted to render plausible the general approach we have been advocating which is to see these patterns as resulting from the effects of TGA rules and Beat Addition

Note next that while it is in principle possible for the GE rules of Beat Movement and Beat Deletion to apply on the second metrical level certain general conditions appear to limit this possibility seshyverely The preservation of the basic beat alignments required by a basic beat rule is guaranteed by the Textual Prominence Preservation Condition Thus two adjacent second-level grid positions will remain in place if both are introduced by the basic beat rules of the language Moreover if a clash is created by the joint effects of a basic beat rule and an application of Beat Addition only the grid position introduced by Beat Addition is susceptible to being deleted In general however clashes on the second metrical level seem to be tolerated quite well There is thus some reason to speculate that Beat Movement and Beat Deletion are simply not applicable on the second metrical level We will give evidence for this in subsequent chapters (In chapter 4 we wil1 propose a condition of Basic Beat Level Integrity (449))

The core of this particular articulation of the metrical grid theory of patterns of stressed and stressless syllables then is a system of rules for constructing the second level of the grid on top of the demibeats aligned with the sequence of syllables contained in a word Universal grammar makes available two distinct sets of rules relevant to metrical level two the basic beat rules (TGA rules) and the rules of grid euphony The choices that a particular language makes from these sets of rules (including the choice of dominance and directionality for Beat Addition)-that is the way it sets these parameters-constitute the linguistic description or grammar of the patterns of stressed and stressless syllables for the language

This core theory of patterns of stressed and stressless syllables does not give expression to lexical idiosyncrasies that may be manifested in

~ i~j


the stress patterns of languages which is as it should be Although it is relatively common in describing individual languages to need to conshysider some syllable of a particular stem or suffix to be inherently stressed this fact should not lead the analyst from the main theme of the investigation which is the understanding of the essential properties of stress systems properties that we are justified in believing to be deshyscribable in terms properly phonological

Lexical idiosyncrasies which can often be expressed as a lexically specified alignment with a basic beat usually occur in languages where the patterns of stress are for the most part (or in large part) the effects of principles of this universally defined core Le basic beat rules and other rules of grid construction Our approach will be to consider that the lexically specified alignments and the alignments introduced by the basic beat rules jointly specify the partial grid to which Beat Addition applies and upon which the filtering function of the PRA is felt

2323 The Third Metrical Level and Above In terms of the metrical grid theory ofprominence patterns main word stress is the alignment of some basic beat of the word with a beat on a higher metrical level In many languages every word will bear a main word stress (except often enough for the class of function words)25 The location of that main word stress may be governed by rule by lexical idiosyncrasies (as is in large part the case in Russian (Halle 1973braquo or by both rule and idiosyncrasy (as for example in English (Liberman and Prince 1977 Hayes 1980 Selkirk 1980braquoln such languages we will say that there is a grammatically governed main word stress

Some languages are reported to have no main word stress (eg Tubatulabal (Swadesh and Voegelin 1939) and 19bo (Green and Igwe 1963 Clark 1978raquo But if the metrical grid and the alignment of the syllables of an utterance with it is a universal as we have suggested then these languages must have an organization of basic beats into beats on the third metrical level or above at least in cases where the utterance contains more than one or two basic beats Though we are in no position to verify it we advance the hypothesis that the seeming lack of main word stress in these languages is to be attributed to the lack of a consistent presence of main stress in words and to the lack of a reliable location for it within a word when it is there This would be the case if prominence on the third metrical level were not guaranteed by a TGA rule with a word-size domain The presence of beats on the third metrical level would be required only to satisfy the PRA which means


that a word would not necessarily have a main stress and if it did it would be introduced by Beat Addition but even then not necessarily if Beat Addition were not obligatory on that level The position of a beat of level three or higher would vary depending on the number and locashytion of basic beats within that word and probably on the number and location of basic beats and beats in neighboring words as well Whether stress systems of this sort exist-assuming of course that they can and do-we leave as an open question

The TGA rules which determine the presence and location of main word stress are the analogue on the third metrical level (or above) to the basic beat rules which determine (in part) the presence and locashytion of beats on the second metrical level Apparently in languages where the location of a strongest beat in the word is TGA-ruJeshygoverned (not lexically (ie idiosyncratically) determined) it tends to be localized toward the ends of the word-most commonly on the first syllable or on the final or penultimate syllable (see Hyman 1977a Hayes 1980)26 The evidence thus points to the existence in the core grammar of word stress of a relatively small number of principles deshytermining the possible locations of main word stress Prince 1981 1983 proposes that there is one basic rule along the following lines

(236) Domain-End Prominence Rule27

The basic beat that is ~ ~tt in the word is aligned with a beat on

a higher metrical level

(The (a) version will be referred to as the Left Domain-End Prominence Rule the (b) version as the Right Domain-End Prominence Rule) The claim is that a language may choose either version and that no other choices are available in core grammar (It will be assumed that the Domain-End Prominence Rule applies even where a cyclic domain has only one basic beat and therefore that main word stress means the alignment of a syllable with a third metrical level at the least Chapters 3 4 and 7 contain explicit arguments for this assumption)

We believe Princes proposal that domain-end prominence rules form the core grammar of primary word stress to be essentially correct though for reasons of space we will not evaluate it right here What is crucial to the metrical grid theory of stress is not so much the character of the rules that define where in the word a beat on the third metrical


level will reliably be located but rather the nature of the patterns of secondary stress that coexist with this main stress on the third metrishycal level (and above) The metrical grid theory of stress which takes the Principle of Rhythmic Alternation as fundamental predicts the existence of a secondary prominence either two or three beats away from the rule-governed main stress And this alternating pattern of secshyondary stress is indeed to be found for instance in Italian and Enshyglish As an example the initial syllables of the English words in (237) are more prominent than the other non-main-stressed beat-aligned syllables

(237) a x b x c x

x x x x x x x x x x x x x x x xx xxxx x x x x x x x x x

reconciliation chimpanzee tintinnabulation

Our hypothesis is that this secondary stress is introduced by Beat Addition which gives the rhythmically regular patterns exhibited These examples show that the Beat Addition applying on the third metshyricallevel is left-dominant (Left-dominance is doubtless the unmarked case for Beat Addition)

Consider now the fact that Beat Addition does not undo the effects of the rule assigning main word stress in English (The Main Stress Rule (MSR) is an instance of the Right Domain-End Prominence Rule see chapter 3) The basic beat rules and the MSR of English derive the following partial grids (again see chapter 3 for details)

(238) a x b x c x

x xx x x x x x x x x xxxx x x x x x x x x x


Beat Addition then applies on the third metrical level But rather than giving either (239a) for example where the added beat and the (forshymerly) main-stressed beat are on a par or (239b) where Beat Addition has applied yet a second time (its structural description being satisfied by the two beats on the third level in (239araquo Beat Addition is accomshypanied by a promotion of the main-stressed beat itself as in (239c)


(239) a

x x x x x x x xxxx

reconciliation

b x x x x x x x x xxxx

reconciliation

c x x x x

recon

x x

x x xxxx

ciliation

The evidence that the main word stress is indeed promoted as in (239c) is the fact that if a pitch accent falls on the word in a non -stress shift environment it will fall on that last beat The general principle govshyerning the assignment of pitch accents to words is that a pitch accent falls on the most prominent syllable of the word (see chapter 5)Thus if the post-Beat Addition representation of reconciliation were (239a) we would expect that (in a non-stress shift environment) the pitch acshycent could fal1 on either the first or the last syllable there being no most prominent one This does not happen If the post-Beat Addishytion representation were (239b) we would expect the pitch accent to fall on the initial beat This does not happen either The conclusion is that (239c) is the appropriate post-Beat Addition representation of reconciliation

But what is the evidence that Beat Addition has applied at all It comes from the behavior of these words under stress shift When Beat Movement applies it throws the most prominent beat back onto the initial beat (There was a) reconciliation ofparties (They heard the) tinshytinnabulation ofbells (You stop these) chimpanzee hijinks This is what would be expected if there had been Beat Addition on that first beat (and promotion of main stress) for that makes the first beat the closest beat to the left of the main stress beat on the next level down and thus the one onto which Beat Movement would place the fourth-level main stress beat Had Beat Addition not applied the second pre-main stress basic beat would have been the first one to the left of main stress on the next level down (see (238raquo Beat Movement therefore would have deshyrived reconcfliation ofparties tintinnabulation of bells chimpanzee hijinks which are ungrammatical

We conclude that Beat Addition does apply in words (in fact necesshysarily so as these examples indicate)28 but that it does not act to over ride the MSR On a higher domain however there is a GE rule Beat Movement which does override the MSR Beat Movement deflects the main stress back to a position it would never have otherwise occupied It is therefore not appropriate to impose the general global condition that GE rules not override the MSR The appropriate generalization


seems to be that within the MSRs own domain of application it may not be undone by a GE rule Thus if we assume that these rules apply in cyclic fashion the Textual Preservation Prominence Condition (225) can be stated and can be given the responsibility of ensuring that the correct patterns are constructed 29

Beat Movement and Beat Deletion can come into play on the third metrical level Adjacent third-level grid positions will be created in Enshyglish by applications of the domain-end prominence rule creating main word stress for example Within the word such a situation will arise only if there have been multiple applications of the MSR on successhysively embedded word-internal cyclic domains Within the phrase a clash between two adjacent third-level beats will arise simply when the main stresses of adjacent words are juxtaposed (with respect to the grid) In both instances that clash may be eliminated by a rule of grid euphony

Kiparsky 1979 argues for example that the stress pattern of a word like expectation is derived from the more basic expect through an apshyplication of the Rhythm Rule -that is stress shift (here Beat Movement) The derivation in grid terms would be as follows

(240) [[ex pect] a tion]

Cycle 1 x

x x DBABBRMSR x x Cycle 2

x x x

x x x a DBA BBR MSR x x x x

x x x x x x

b BM x x x x

Examples of Beat Movement on the phrase have already appeared in section 22

In English Beat Movement to the left takes precedence over Beat Deletion thus guaranteeing that prominence will be preserved on that level of the grid albeit in a location different from its original one with respect to the text But Beat Movement in English appears to be asymmetrical (Given the core theory approach this means that the


grammar of English chooses only the (223b) left movement version of Beat Movement from the universal set available) As Liberman and Prince point out the compound sports contest with greatest prominence on sports as in (241) does not become (242) by Beat Movement

(241) (242) x x x x x x x x x x x x x x x x x x

sports contest sports contest

If anything the underlying (241) simply undergoes Beat Deletion to become (243)

(243) x x x x x x x

sports contest

And Beat Addition not obligatory above the level of the word will not act to restore the prominence on con Thus Beat Deletion in English applies where Beat Movement does not We would suggest that Beat Deletion too is asymmetrical in other words that the grammar of Enshyglish has chosen (224b) from the universally available set (see section 422)

The TGA rules of English that apply on syntactic domains larger than the word are the Compound Rule and the Nuclear Stress Rule (NSR) The English Compound Rule assigns prominence on the fourth metrical level or above to the most prominent third-level beat contained within the first (or leftmost) immediate constituent of the compound In simishylar fashion the NSR assigns rhythmic prominence at the right extreme of a phrase We are inclined to make the rather strong hypothesis that the TGA rules applying on all such syntactic domains are domain-end prominence rules in this particular sense (See the discussion in chapshyter 4) The NSR will ensure that the text of a sentence like (244) will receive the partial grid alignment of (245) (ignoring the metrical levels below the three of main word stress)


(244) [It was [organized model [a gallon

(245)

x x x x x

It was organized on the model of a gallon of worms

This is what would be derived if Beat Addition were not to apply in the course of the cyclic grid construction It could apply however elimshyinating the rhythmic lapses The other possible outcomes for (244) therefore are (246) and (247)

(246) x

x x x x x x


(247) x

x x x x x x


Note that any application of Beat Addition requires the promotion of the main stress required by the domain-end prominence rule the NSR At the phrase level then as at the word level the two sorts of grid construction rules interact in the same way bringing about rhythmic patterns that respect the requirements of the text

2324 Summary In proceeding level by level in the preceding exposition we have sought to give plausibility to our claim that the patterns of rhythmic prominence in language can be insightfully charshyacterized by a set of quite general grid-based rhythmic principles the rules of grid euphony in conjunction with a set of syntactic-structureshybased principles the texHo-grid alignment rules Our argument for the necessity of viewing the metrical grid as the representation of promshyinence patterns in terms of which the core theory of prominence patterns is to be cast relies on showing that real generalizations are captured in this way that none are (systematically) lost and that no other theory can do as well In the following chapters we will further elaborate this theory and the particular analysis of English stress pat-


terns on which our argument is partly based Where necessary we will make the appropriate comparisons with alternative theories

So far we have said nothing about the ordering of the various sorts of rules in the derivation of syllable-to-grid alignments of sentences In fact there is not much to be said for we will consider most of the ordering relations to be intrinsically defined Demibeat Alignment must precede anything else the basic beat rules must precede Beat Addition on the second metrical level the main stress rules must precede Beat Addition on the third metrical level etc Another general ordering principle that suggests itself is that the alignment should proceed level by level from the lowest to the highest The facts of many languages are consistent with this level ordering In Cairene Classical Arabic or in Italian for example the assignment of beats at the third metrical level in both cases carried out by the Right Domain-End Prominence Rule presupposes a prior assignment of basic beats on the second metshyrical level Perhaps then there is in fact nothing to be said in the grammar of a language about ordering in the building of the syllableshyto-grid alignments

There are languages though such as Russian in which alignment on the third metrical level precedes organization into basic beats on the second level In Russian the locus of main stress (third-level beat alignment) is in part lexically (morphologically) determined (Halle 1973a) and that lexically specified alignment with the third (and hence second) metrical level serves as the axis around which organization into beats on the second metrical level is defined The pretonic (pre -main stress) syllable in Russian is always associated with a beat on the secshyond metrical level and the syllables preceding it appear in an alternatshying pattern (Karcevskij 1931) It would be interesting to know whether all divergences from level ordering appear in cases like Russian where the specification of the upper level alignment is a matter of lexishycal idiosyncrasy If so then it will be possible to retain the idea that the ordering of grid construction rules by metrical level (from lower to higher) is the unmarked case the state of affairs when all proceeds according to principles that are phonological in character


22 The Rhythmic Nature of Speech

Scholars working on the phonetics ofEnglish have long recognized that rhythmic organization at what we will call the lower levels plays an important role in the description of speech 1 A major insight of these investigations is that the quantities (durations) of the syllables of a senshytence are determined by the sentences rhythmic properties According

to D Jones 1964sect886 for example Vowel length depends to a considerable extent [in English on the rhythm of the sentence There is a strong tendency in connected speech to make stressed syllables follow each other as nearly as possible at equal distances For D Jones 1964sect888-890 the musical notation in (2la) and (21b) represents the fact that in (21a) ei ai are something like twice as long in the second sequence as they are in the first and that in (21b) the i in scene is considerably longer than the i in scenery

(21) a n

eitin oi~ FJ

naintin twenti vs )

eit ) )

nain ten

eighteen nineteen twenty eight nine ten

b )i ) m )rnr ()~ sin w~z bjUltafi vs 6a sinari waz bjutafl The scene was beautiful The scenery was beautiful

Pike 194534 makes a not dissimilar observation about American Enshyglish He points out that the space of time elapsing between the stressed syllables man and here is roughly the same in The man is here and The managers here with the result that in the latter the syllables are crushed together (pronounced very rapidly) in order to be fitted

in Building on the insights of Jones and others Abercrombie 1964 laid

the foundation for further work oil the topic by positing a unit of analyshysis the foot (not to be confused with the foot of metrical phonology) in terms of which he argued the rhythmic properties of syllables may be

explained English utterances may be considered as being divided by the isoshychronous beat of the stress pulse into feet of (approximately) even


length Each foot starts with a stress and contains everything that folshylows that stress up to but not including the next stress This is the house that lack built has therefore four feet and they can be most conveniently represented by the use of vertical lines

This is thelhouse thatlJacklbuilt The quantity of any syllable is a proportion of the total length of the foot within which the syllable occurs and it is relative to the quantity of any other syllable in the foot (1964217)

The claim being made then is that English utterances can be charshyacterized in rhythmic terms as consisting of a sequence of isochronous pulses This claim about isochrony in English speech has engendered considerable debate (see Lehiste 1980 for a review) 2 We feel that much of this debate and the apparent lack of resolution on the question stems from a failure to understand rhythm and the isochrony of its basic pulses in sufficiently abstract terms With Liberman Lehiste and others we claim a psychological reality for rhythmic organization in both the production and the perception of speech but acknowledge that the ideal isochronicity this organization presupposes may not alshyways reveal itself in easily measurable terms in the acoustic signal

We may liken the metrical grid alignment of a sentence- the linguisshytic representation of the sentences rhythmic structure-to a musical score It must be recognized that a musical score or a metrical grid alignment though grounded in its very conception by the rhythmic cashypacities of the human organism (the rules for defining possible scores or metrical grid alignments embodying as they do universals of rhythmic organization) is but an abstract scheme The same score may be intershypreted in many different ways by the same or different performers though of course there are limits within which variation in interpretashytion is confined if the score is to remain recognizable that is if the score is to be considered to have been realized in the performance A distinction must therefore be made between the score and its interpreshytation This is the distinction between langue and parole or between competence and performance (using the latter term now in its technical sense) For linguistic patterns of prominence this is the distinction between the metrical grid alignment of the sentence and its phonetic implementation In measuring for isochrony one is measuring only the performance of the score and not the abstract patterning that makes it up We give full credence therefore to the impression of isochronyshyto the impression of rhythm-as revealing something about how the mind grasps the organization of speech in time 3 It is what the mind




(22) x x

The managers here


(23) x x






(24)








(25)






(26)









(28)









M~v CL L



Figure 21

x x X




Figure 22







(210)

a X b x X ~ ~

X X x

x x



x x

x x












(213) 000

xxx




(214)





(215) x x x x x


(216) x xx

x xx x

(217) x




(218) x

x x x xx x x x xx




(219) (220) x x


























(222)

Beat Addition



(223)

Beat Movement

x x x x x x




(224) Beat Deletion

x x x x x c x x x


b x x = x x xx=xx































(230) x x x x x x x

(J (J (J (J (J (J






(231) odd even

x x x x

x x x

x x x x x x x x x



(232) odd even





(233)





b Beat Addition
















~ i~j
















(237) a x b x c x





(238) a x b x c x





(239) a

x x x x x x x xxxx

reconciliation


reconciliation

c x x x x

recon

x x

x x xxxx

ciliation









Cycle 1 x


x x x


x x x x x x

b BM x x x x








(243) x x x x x x x

sports contest





(245)

x x x x x



(246) x

x x x x x x


(247) x

x x x x x x











(22) x x

The managers here


(23) x x






(24)








(25)






(26)









(28)









M~v CL L



Figure 21

x x X




Figure 22







(210)

a X b x X ~ ~

X X x

x x



x x

x x












(213) 000

xxx




(214)





(215) x x x x x


(216) x xx

x xx x

(217) x




(218) x

x x x xx x x x xx




(219) (220) x x


























(222)

Beat Addition



(223)

Beat Movement

x x x x x x




(224) Beat Deletion

x x x x x c x x x


b x x = x x xx=xx































(230) x x x x x x x

(J (J (J (J (J (J






(231) odd even

x x x x

x x x

x x x x x x x x x



(232) odd even





(233)





b Beat Addition
















~ i~j
















(237) a x b x c x





(238) a x b x c x





(239) a

x x x x x x x xxxx

reconciliation


reconciliation

c x x x x

recon

x x

x x xxxx

ciliation









Cycle 1 x


x x x


x x x x x x

b BM x x x x








(243) x x x x x x x

sports contest





(245)

x x x x x



(246) x

x x x x x x


(247) x

x x x x x x












(25)






(26)









(28)









M~v CL L



Figure 21

x x X




Figure 22







(210)

a X b x X ~ ~

X X x

x x



x x

x x












(213) 000

xxx




(214)





(215) x x x x x


(216) x xx

x xx x

(217) x




(218) x

x x x xx x x x xx




(219) (220) x x


























(222)

Beat Addition



(223)

Beat Movement

x x x x x x




(224) Beat Deletion

x x x x x c x x x


b x x = x x xx=xx































(230) x x x x x x x

(J (J (J (J (J (J






(231) odd even

x x x x

x x x

x x x x x x x x x



(232) odd even





(233)





b Beat Addition
















~ i~j
















(237) a x b x c x





(238) a x b x c x





(239) a

x x x x x x x xxxx

reconciliation


reconciliation

c x x x x

recon

x x

x x xxxx

ciliation









Cycle 1 x


x x x


x x x x x x

b BM x x x x








(243) x x x x x x x

sports contest





(245)

x x x x x



(246) x

x x x x x x


(247) x

x x x x x x











(28)









M~v CL L



Figure 21

x x X




Figure 22







(210)

a X b x X ~ ~

X X x

x x



x x

x x












(213) 000

xxx




(214)





(215) x x x x x


(216) x xx

x xx x

(217) x




(218) x

x x x xx x x x xx




(219) (220) x x


























(222)

Beat Addition



(223)

Beat Movement

x x x x x x




(224) Beat Deletion

x x x x x c x x x


b x x = x x xx=xx































(230) x x x x x x x

(J (J (J (J (J (J






(231) odd even

x x x x

x x x

x x x x x x x x x



(232) odd even





(233)





b Beat Addition
















~ i~j
















(237) a x b x c x





(238) a x b x c x





(239) a

x x x x x x x xxxx

reconciliation


reconciliation

c x x x x

recon

x x

x x xxxx

ciliation









Cycle 1 x


x x x


x x x x x x

b BM x x x x








(243) x x x x x x x

sports contest





(245)

x x x x x



(246) x

x x x x x x


(247) x

x x x x x x









M~v CL L



Figure 21

x x X




Figure 22







(210)

a X b x X ~ ~

X X x

x x



x x

x x












(213) 000

xxx




(214)





(215) x x x x x


(216) x xx

x xx x

(217) x




(218) x

x x x xx x x x xx




(219) (220) x x


























(222)

Beat Addition



(223)

Beat Movement

x x x x x x




(224) Beat Deletion

x x x x x c x x x


b x x = x x xx=xx































(230) x x x x x x x

(J (J (J (J (J (J






(231) odd even

x x x x

x x x

x x x x x x x x x



(232) odd even





(233)





b Beat Addition
















~ i~j
















(237) a x b x c x





(238) a x b x c x





(239) a

x x x x x x x xxxx

reconciliation


reconciliation

c x x x x

recon

x x

x x xxxx

ciliation









Cycle 1 x


x x x


x x x x x x

b BM x x x x








(243) x x x x x x x

sports contest





(245)

x x x x x



(246) x

x x x x x x


(247) x

x x x x x x


















(213) 000

xxx




(214)





(215) x x x x x


(216) x xx

x xx x

(217) x




(218) x

x x x xx x x x xx




(219) (220) x x


























(222)

Beat Addition



(223)

Beat Movement

x x x x x x




(224) Beat Deletion

x x x x x c x x x


b x x = x x xx=xx































(230) x x x x x x x

(J (J (J (J (J (J






(231) odd even

x x x x

x x x

x x x x x x x x x



(232) odd even





(233)





b Beat Addition
















~ i~j
















(237) a x b x c x





(238) a x b x c x





(239) a

x x x x x x x xxxx

reconciliation


reconciliation

c x x x x

recon

x x

x x xxxx

ciliation









Cycle 1 x


x x x


x x x x x x

b BM x x x x








(243) x x x x x x x

sports contest





(245)

x x x x x



(246) x

x x x x x x


(247) x

x x x x x x











(215) x x x x x


(216) x xx

x xx x

(217) x




(218) x

x x x xx x x x xx




(219) (220) x x


























(222)

Beat Addition



(223)

Beat Movement

x x x x x x




(224) Beat Deletion

x x x x x c x x x


b x x = x x xx=xx































(230) x x x x x x x

(J (J (J (J (J (J






(231) odd even

x x x x

x x x

x x x x x x x x x



(232) odd even





(233)





b Beat Addition
















~ i~j
















(237) a x b x c x





(238) a x b x c x





(239) a

x x x x x x x xxxx

reconciliation


reconciliation

c x x x x

recon

x x

x x xxxx

ciliation









Cycle 1 x


x x x


x x x x x x

b BM x x x x








(243) x x x x x x x

sports contest





(245)

x x x x x



(246) x

x x x x x x


(247) x

x x x x x x




























(222)

Beat Addition



(223)

Beat Movement

x x x x x x




(224) Beat Deletion

x x x x x c x x x


b x x = x x xx=xx































(230) x x x x x x x

(J (J (J (J (J (J






(231) odd even

x x x x

x x x

x x x x x x x x x



(232) odd even





(233)





b Beat Addition
















~ i~j
















(237) a x b x c x





(238) a x b x c x





(239) a

x x x x x x x xxxx

reconciliation


reconciliation

c x x x x

recon

x x

x x xxxx

ciliation









Cycle 1 x


x x x


x x x x x x

b BM x x x x








(243) x x x x x x x

sports contest





(245)

x x x x x



(246) x

x x x x x x


(247) x

x x x x x x









(224) Beat Deletion

x x x x x c x x x


b x x = x x xx=xx































(230) x x x x x x x

(J (J (J (J (J (J






(231) odd even

x x x x

x x x

x x x x x x x x x



(232) odd even





(233)





b Beat Addition
















~ i~j
















(237) a x b x c x





(238) a x b x c x





(239) a

x x x x x x x xxxx

reconciliation


reconciliation

c x x x x

recon

x x

x x xxxx

ciliation









Cycle 1 x


x x x


x x x x x x

b BM x x x x








(243) x x x x x x x

sports contest





(245)

x x x x x



(246) x

x x x x x x


(247) x

x x x x x x























(230) x x x x x x x

(J (J (J (J (J (J






(231) odd even

x x x x

x x x

x x x x x x x x x



(232) odd even





(233)





b Beat Addition
















~ i~j
















(237) a x b x c x





(238) a x b x c x





(239) a

x x x x x x x xxxx

reconciliation


reconciliation

c x x x x

recon

x x

x x xxxx

ciliation









Cycle 1 x


x x x


x x x x x x

b BM x x x x








(243) x x x x x x x

sports contest





(245)

x x x x x



(246) x

x x x x x x


(247) x

x x x x x x










(233)





b Beat Addition
















~ i~j
















(237) a x b x c x





(238) a x b x c x





(239) a

x x x x x x x xxxx

reconciliation


reconciliation

c x x x x

recon

x x

x x xxxx

ciliation









Cycle 1 x


x x x


x x x x x x

b BM x x x x








(243) x x x x x x x

sports contest





(245)

x x x x x



(246) x

x x x x x x


(247) x

x x x x x x























(237) a x b x c x





(238) a x b x c x





(239) a

x x x x x x x xxxx

reconciliation


reconciliation

c x x x x

recon

x x

x x xxxx

ciliation









Cycle 1 x


x x x


x x x x x x

b BM x x x x








(243) x x x x x x x

sports contest





(245)

x x x x x



(246) x

x x x x x x


(247) x

x x x x x x













Cycle 1 x


x x x


x x x x x x

b BM x x x x








(243) x x x x x x x

sports contest





(245)

x x x x x



(246) x

x x x x x x


(247) x

x x x x x x










(245)

x x x x x



(246) x

x x x x x x


(247) x

x x x x x x








Chapter 2 Rhythmic Patterns in Language It

Documents

Transcript of Chapter 2 Rhythmic Patterns in Language It